Voice Quality Sonification — User Guide

Maps local jitter and shimmer trajectories to two continuously changing spectral bands, turning measured voice-quality instability into an audible time-varying timbral transformation.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 1.2 (2026) Technique: Voice analysis → parameter-mapped spectral filtering License: MIT License
Contents:

What this does

Voice Quality Sonification is a data-to-sound processor that measures local jitter and shimmer across a voice recording and uses those trajectories to move two spectral pass bands over time. Jitter and shimmer therefore become control signals for a timbral transformation: one trajectory moves a lower spectral region and the other moves a higher region.

The processor preserves the source recording as the sound material. At each analysis point it creates a pair of filtered versions of the full source, then crossfades continuously between neighboring filter states. The result follows the measured voice-quality trajectory without hard segment boundaries.

Important interpretation: the two bands are user-defined spectral regions. They may be placed in formant-like frequency ranges, but the script does not estimate formants, track formants, or synthesize a vocal-tract model. Its purpose is creative sonification of measured perturbation data through time-varying spectral filtering.

Processing pipeline

  1. Prepare an analysis source. Mono input is analyzed directly. Multichannel input is averaged to mono; if that average is nearly cancelled, channel 1 is used instead.
  2. Sample the recording at evenly spaced analysis points. Each point uses a rectangular window centered on its analysis time.
  3. Measure local perturbation. Praat estimates pulses and measures local jitter and local shimmer in each valid window.
  4. Normalize each trajectory within the file. Valid jitter values are min–max scaled to 0–1, and shimmer is normalized independently.
  5. Repair missing control points. An invalid window inherits the nearest valid normalized value for that metric. A metric that is undefined throughout remains neutral at 0.5.
  6. Map controls to two spectral bands. The normalized values shift the lower and higher pass-band ranges upward or downward by the selected mapping depth.
  7. Create filter states. At every analysis point, the complete multichannel source is filtered into a low spectral band and a high spectral band; the two bands are summed.
  8. Interpolate through time. Adjacent filter-state signals are linearly crossfaded so the weights sum to one throughout the file.
  9. Mix dry and wet. The time-varying filtered signal is blended with the original according to Dry_wet_mix.
  10. Apply safety attenuation if required. Safety_peak only turns the result down when its peak exceeds the requested ceiling.

Quick start

  1. Select exactly one Sound object in Praat.
  2. Run Voice_Quality_Sonification.praat.
  3. Start with Moderate Effect.
  4. Use clearly periodic or voiced material so that local pulse measurements are available.
  5. Leave the default low and high bands at 280–900 Hz and 900–2500 Hz for a broad, voice-centered spectral mapping, or set your own regions.
  6. Use Dry_wet_mix to control how strongly the sonification replaces the original spectrum.
  7. Enable Draw_analysis to inspect the measured controls and mapped band trajectories.
Reading the result: the audible movement represents the relative trajectory within this recording. A normalized value of 1 means the largest valid measurement found in this file, not an absolute clinical reference value.

Jitter & shimmer analysis

Analysis grid

Num_windows analysis centers are distributed evenly from half a window after the beginning to half a window before the end. The extraction window is rectangular so that an amplitude taper does not artificially create shimmer.

analysisTime[1] = Window_size_s / 2
analysisTime[last] = duration - Window_size_s / 2
intermediate points = evenly spaced between them

Pulse and perturbation measurements

Each window is converted to a Praat cross-correlation pitch representation, aligned to the waveform with Sound & Pitch: To PointProcess (cc), then queried with Get jitter (local) and Get shimmer (local). A window needs at least four pulse points before the script attempts either measurement.

Internal settingValue
Pitch range75–600 Hz
Period floor1/600 s
Period ceiling1/75 s
Maximum period factor1.3
Maximum amplitude factor1.6

Within-file normalization

Jitter and shimmer are normalized separately using only their valid windows:

control = (value - minimum_valid_value) / (maximum_valid_value - minimum_valid_value)

If all valid values of one metric are equal, that metric is held at 0.5. Invalid windows copy the nearest valid control value. If jitter is unavailable throughout, only the jitter-derived control remains neutral; the same rule applies independently to shimmer. The script stops only when neither metric has any valid measurement.

Mapping & filtering

Control-to-frequency mapping

A normalized control of 0.5 leaves its base band unchanged. At the normal mapping direction, 0 shifts the band downward by the selected depth and 1 shifts it upward by the same percentage.

shift = 1 + 2 × (control - 0.5) × mapping_depth
shifted_low_edge = base_low_edge × shift
shifted_high_edge = base_high_edge × shift

For example, a depth of 0.15 produces a frequency multiplier from 0.85 to 1.15 across the normalized trajectory. Invert Direction reverses that relationship. Frequency edges are clamped to safe limits below Nyquist.

Four mapping modes

Mapping modeLow bandHigh band
Jitter → low, Shimmer → highJitterShimmer
Shimmer → low, Jitter → highShimmerJitter
Jitter/Shimmer inverted directionJitter, invertedShimmer, inverted
Swapped + inverted directionShimmer, invertedJitter, inverted

Dual-band state

Each state consists of two Praat pass Hann band filters applied to the full source. The two filtered bands are summed. Filter_transition_Hz is the spectral smoothing width at the edges of each pass band; it is independent of the temporal interpolation between analysis states.

Temporal motion is created by a linear crossfade between the complete filtered signals of adjacent states. Before the first analysis center the first state is held; after the last center the final state is held.

Presets

PresetWindowsWindowMappingJitter depthShimmer depth
CustomUserUserUserUserUser
Subtle Variation6250 msJitter → low / Shimmer → high0.080.06
Moderate Effect8200 msJitter → low / Shimmer → high0.150.12
Extreme Mapping12150 msJitter → low / Shimmer → high0.300.25
Swap Jitter/Shimmer8200 msShimmer → low / Jitter → high0.150.12
Invert Direction8200 msJitter → low / Shimmer → high, inverted0.150.12
Preset scope: presets override the analysis count, analysis-window length, mapping mode and the two mapping depths shown above. Base band ranges, spectral transition width, Dry/Wet, Safety peak, visualization and playback remain whatever is set in the form.

Parameters

ParameterDefaultBehavior
Num_windows8Number of local measurements/control points. Internally clamped to 2–64.
Window_size_s0.2 sRectangular measurement window. Minimum 0.05 s; the source must be at least this long.
Mapping_modeJitter → low, Shimmer → highSelects assignment and/or directional inversion.
Jitter_mapping_depth0.15Maximum proportional shift of the jitter-controlled band. Magnitude is clamped to 0–0.90.
Shimmer_mapping_depth0.12Maximum proportional shift of the shimmer-controlled band. Magnitude is clamped to 0–0.90.
Low_band_low_Hz / Low_band_high_Hz280 / 900 HzBase edges of the lower pass band.
High_band_low_Hz / High_band_high_Hz900 / 2500 HzBase edges of the higher pass band.
Filter_transition_Hz100 HzHann smoothing width at each pass-band edge; minimum 1 Hz.
Dry_wet_mix1.00 = dry source, 1 = time-varying dual-band output. Clamped to 0–1.
Safety_peak0.99Attenuation-only output ceiling. 0 disables the ceiling.
Draw_analysisOnDraw the AudioTools diagnostic visualization.
Play_resultOnPlay the final Sound after processing.

Input & output behavior

Visualization

When Draw_analysis is enabled, v1.2 draws four diagnostic regions plus a summary strip:

The visualization is diagnostic only. In testing, the same settings produced sample-identical audio with drawing enabled and disabled.

Notes & limitations

Further Reading